Can UV absorbance rapidly estimate the chlorine demand in wash water during fresh-cut produce washing processes?

•Chlorine demand (CLD) in vegetable wash water was estimated with UVA measurements.•UVA changed with organic load and degree of chlorination (remaining CLD of water).•Predicting CLD required UVA measurement at two wavelengths.•A prediction model was made for CLD of Greenleaf lettuce wash water and validated. Free chlorine is used in industrial fresh-cut produce washing to avoid cross-contamination from pathogenic and spoilage microorganisms, although chlorine dosing typically depends on feedback control. Control of free chlorine levels in fresh-cut produce wash water could be improved if chlorine demand (CLD) could be determined real-time, during processing. Previous research has shown that the CLD of non-chlorinated fresh produce wash water (CLDmax) correlates with UV absorbance (UVA) at 254 nm (UVA254). The goal of this study was to estimate CLD for produce wash conditions that are in-progress, i.e., when the chlorine concentration in water partially meets the CLD, as is the case during industrial, continuous produce washing. This was done for cabbage, carrot, green leaf lettuce and onion. UVA changed with both CLDmax and remaining CLD. Two wavelengths were necessary to predict the CLD:UVAmin, which changed minimally due to chlorination and had maximum correlation with CLDmax and UVAmax. The CLDmax and UVAmax changed maximally with chlorination and had maximum correlation with the fraction of the remaining CLD. Results showed that UVAmin and UVAmax were between 240–290 nm, and the exact wavelength depended on the vegetable. However, free chlorine itself influences UVA, and at a residual above 25 mg/L the chlorine interfered with the estimation of CLD. A case study on green leaf lettuce showed that CLD can be predicted by a model of the form f(UVAmin) x g(UVAmax /UVAmin). Using external validation data, optimal predictability of the model was obtained when both f and g were expressed as quadratic equations (SD/RMSE = 3.55; R² = 0.93). The described UVA method for predicting CLD shows promise for online application. Further studies should incorporate the possible variability in crop composition as well as other possible interferences with the UVA signal.